Introduction

modiff looks for differentially methylated loci.

By default it does a two group tests by using Bayes factors with a binomial model for the counts. Other models can be specified.

Usage is as folllows:

modiff [--model][--groups] FILE

where FILE can either be:

• A TAB-separated file containing modified bases counts. The columns are contig and genomic coordinate followed by depth and number of modified bases per each sample. For example with 2 samples an input line looks like the following:

  chr1\t10471\t20\t7\t23\t19

  In this case modiff can read directly from the stdin if the file name is -.

or

• a bgzip compressed, tabix indexed TAB-separated file, organized as above. the tabix format is automatically recognized by modiff.

Options

--model choose the model used to compare the two groups. (see below)

--groups specify the which sample belongs to which group.

Unless the --groups option is set, the first half of the count columns are assumed to belong to group 0 and the remaining half to group 1; otherwise groups can be used to assign samples to groups, eg with 4 samples one can call

modiff --groups 0:1:1:0 samples.txt

Models

• logreg corresponds to running a binomial logistic regression with an intercept term and a group term, and testing the significance of the group term

• lr corresponds to computing the log likelihood ratio between a model with two parameters (one per group) and a model with a single parameter (across groups). Highest values of lr give credibility to the two parameters model.

• bf computes the log of the Bayes factor between a two parameter model and a one parameter model (integrating over the parameters).

if no model is specified through --model the default is lr.

output format

modiff produces an output file compliant with the BEDn specifications https://samtools.github.io/hts-specs/BEDv1.pdf (0-based, half open coordinate system).

The output format changes depending on the model used, but the first few columns are always the same.

column	quantity	comment
1	contig/chromosome	reference sequence
2	start position	(inclusive, 0-based)
3	end position	(exclusive, 0-based)
4	mean methylation by group	comma separated
5	depth by group	comma separated

in the other model the other columns are to be documented but the last column always contain an indicator of strenght of effect or significance (pvalue or Bayes factor).

Contact

github issues and/or pull requests. email to emanuele dot raineri at cnag dot eu